About ConvertMP3toMIDI

What this site is

ConvertMP3toMIDI is a free audio-to-MIDI converter. You drop in an audio or video file, and the page writes out a standard MIDI file (.mid) that you can open in any DAW or notation program — Ableton, FL Studio, Logic, GarageBand, MuseScore and the rest.

MP3, WAV, M4A, FLAC, OGG, AIFF, Opus and WebM are all accepted, and the audio track of MP4, M4V or MOV files is read too.

Who made it

This site is made by ConvertMP3toMIDI. The note transcription itself is not ours to claim: it runs on GAME (Generative Adaptive MIDI Extractor), the open-source singing-voice transcription model from OpenVPI, exported to ONNX and executed on your own device with ONNX Runtime Web. GAME's code is MIT-licensed; the pretrained model weights are released under CC BY-NC-SA 4.0, so this site uses them for non-commercial purposes only.

Why it works the way it does

Two things shaped it. First, your audio should not have to travel anywhere for you to get a MIDI file, so the whole pipeline — decoding, analysis, MIDI writing — runs inside your browser tab. Nothing is uploaded, and there is nothing to sign up for. The model itself is about 51 MB; it is fetched from a pinned revision of the official mirror, verified by checksum, and then cached, so the download happens once. Second, transcription is estimation, not truth. Rather than print an accuracy number nobody can verify, the page shows you the result before you download it: note count, pitch range, and a piano roll you can read at a glance. If it looks wrong, you know before you leave the page.

What it is good at, and what it is not

GAME is built for singing voice and writes a single melodic line. That is a deliberate trade: on a vocal or a lead melody it avoids the harmonic ghost notes that pitch-tracking models produce, and on a drum-only recording it writes nothing at all rather than inventing a bass part out of the kick. It will not hand you chords, bass and drums as separate parts from a dense full mix — it reads the melody it can hear and leaves the rest.

Two behaviours are worth knowing before you rely on it. On a drum-only file the output is empty, by design; that is the tool telling you a melody transcriber is the wrong instrument for percussion. And on a full mix with drums, the drums do not add wrong pitches — they re-trigger the melody's own notes, so the output can contain several times more notes than the melody plays. In our own measurements the same melody line produced 17 notes on its own and 74 with a drum kit underneath it, with the pitch range unchanged. Stripping the drums before converting avoids both problems.

What it costs

Nothing. There is no account, no email gate, no daily limit, and no payment. Because the model runs on your device rather than on a server, there is no per-conversion cost to hand on to you.

Getting in touch

Questions, bug reports and corrections are welcome — use the contact page. If a file converts badly and you can share it, that is the fastest way to make the tool better.